Get Started
Research questionHow can speech recognition avoid hallucinated transcripts on non-speech without degrading genuine speech?Generative speech recognition decoders can produce fluent text even when audio contains little or no speech. Suppressing these outputs is difficult because interventions may also reject or distort genuine speech.
AI
Audio & Speech
Audio & Speech Processing
Machine Learning
Latest papersRecent research connected to this question, newest first.Reducing Hallucinated Transcripts in Whisper via Hallucination Space ProjectionThe evidence concerns Whisper and a training-free inference-time intervention calibrated on non-speech data. Results are reported on non-speech benchmarks and LibriSpeech, measuring hallucination reduction alongside word error and false-rejection effects.research paper · Sep 3, 2026
Related questions
How can large vision-language models reduce visual hallucinations without extra training or decoding passes?How can speech deepfake detectors focus on synthesis artifacts while generalizing to unseen speakers?How can speech enhancement adapt to mismatched deployment acoustics without labeled target audio?How can black-box LLM APIs detect hallucinations without trusted context while limiting false alarms?
Home
Topics
Search
Library